OpenAI made GPT-Live-1.1 available in its API on November 19, a new model for full-duplex voice agents that listen and speak at the same time. The release adds more natural and consistent accents across languages, a standby option intended to lower idle costs, and more reliable task completion. It is the plumbing for a world in which talking to a business means talking to an agent, and the agent talks back without waiting for you to finish.
GPT-Live-1.1, model name gpt-live-1.1, is OpenAI voice agent model for the API, built for full-duplex conversation over WebRTC or WebSocket transports. Full-duplex means the model listens and speaks at the same time, the way people do, instead of waiting for a silence before it replies. It can hear an interruption, pick up a correction mid-sentence, and keep the conversation moving.
The model also brings more natural and consistent accents across languages, which matters for businesses that serve customers in more than one country. An agent that sounds right in each language is an agent customers will actually talk to, and consistency across languages is what makes a global deployment feel like one product instead of many.
The release adds a standby option intended to lower idle costs, which is the kind of change developers notice immediately. A voice agent spends much of its life waiting: on hold, listening for a caller to return, or sitting between turns. Paying full price for that waiting time made long calls expensive, and the standby mode prices the idle minutes lower.
OpenAI bills the model per second at 5 cents per active minute and 2 cents per minute on WebRTC sessions while inference is paused in standby. That is a meaningful saving on calls with long pauses, such as customer support calls where the agent waits while the caller checks something. Pricing that rewards efficient agents will produce more efficient agents.
Voice agents are already answering phones for customer support, taking orders, booking appointments and qualifying leads, and the quality of the conversation is the limit on how far they spread. A model that can handle interruptions, keep its accent consistent and stay affordable while waiting is a model that can take on harder calls, including the ones where the customer is frustrated, confused or in a hurry.
The same technology powers the voice mode in consumer assistants, which means improvements here flow both ways. A better API model makes a better ChatGPT voice mode, and the lessons from millions of consumer conversations make the API model better. The two products improve each other.
For search, voice agents are another surface where answers are spoken rather than listed. A customer who asks a voice agent about a product gets an answer drawn from somewhere, and the somewhere that gets cited is the page with the clearest facts. Voice search has always rewarded direct answers, and full-duplex agents reward them even more, because there is no screen to show a list of links.
For SEOs, the practical step is the same as for every conversational surface: publish facts that can be spoken in one sentence, keep them current, and make sure the page that holds them is the page an agent would retrieve. The agent is the new searcher, and it listens for clarity.
A customer who asks a voice agent about a product gets an answer drawn from somewhere, and the somewhere that gets cited is the page with the clearest facts. Voice search has always rewarded direct answers, and full-duplex agents reward them even more, because there is no screen to show a list of links.
Publish facts that can be spoken in one sentence, keep them current, and make sure the page that holds them is the page an agent would retrieve. The agent is the new searcher, and it listens for clarity.
Sources: OpenAI, November 19, 2026.
Related reading on this site: November Core Update Volatility Stays Muted as Rollout Nears Its Second Week, Gemini 4 Hydrogen Is Now Live in Google Search Worldwide, Google Moved AI Overviews to the Side for Brand Searches, and Knowledge Panels Are Fading, OpenAI Launches ChatGPT for Sales, a Plugin That Wants to Run Your Whole Sales Workflow, Perplexity Computer Multiagent Puts a Team of 16 AI Agents on One Task, and AI Overviews Are Taking Over the Map Pack, Semrush Finds.
Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.
Start free trial